2 Doctoral Researchers in Reinforcement Learning

We are currently seeking two highly motivated doctoral researchers to join the at Aalto University’s School of Electrical Engineering for an ERC Starting Grant-funded project focused on continual reinforcement learning for real-world systems. The grant provides a stable, well-resourced four-year research environment to tackle a problem that classical RL theory is not built for: real systems don’t reset, failures can be irreversible, and the real world keeps changing.

Standard RL finds policies by optimizing over many hypothetical futures. Under dynamics, such an average may differ arbitrarily from what the individual agent experiences as it lives out one trajectory over time. Furthermore, RL typically seeks a time-invariant policy that cannot adapt to a changing environment. This project addresses both challenges simultaneously, combining trajectory-centric objectives with adaptive, context-dependent policies. This enables autonomous vehicles to adapt to seasonal changes over years, manufacturing systems to evolve with production demands without downtime, and medical monitoring to personalize to patients over decades.

The successful candidates will develop fundamental theory and practical algorithms that move reinforcement learning closer to real-world systems. Position 1 will focus on trajectory-centric optimization, Position 2 on adaptive policies, and both will work jointly to integrate the two approaches into a single RL algorithm. While the work is mainly methodological and theoretically focused, our Aalto robot lab offers the possibility to evaluate the algorithms through practical experiments, e.g., on robot arms or quadruped robots.

Research focus

This project sits at the intersection of two problems: connecting ergodicity theory to reinforcement learning is a very recent development, and enabling RL agents to safely detect and adapt to a changing environment is a timely and active research area. Combining trajectory-centric objectives and adaptive, context-aware policies is new territory. The two doctoral researchers will drive this forward from these two complementary angles and will work closely together to integrate both into a single algorithm:

• Developing theory for trajectory-centric stochastic optimization under non-ergodic dynamics;
• Implementing practical RL algorithms that optimize long-term performance of individual agents;
• Developing change detection algorithms to infer when a policy is no longer valid and needs to be adapted;
• Implementing efficient policy-adaptation algorithms with safety guarantees;
• Integrating those advances into state-of-the-art RL algorithms;
• Validating the algorithms in relevant simulation environments and hardware experiments.

Requirements

Eligible candidates should have

• Master’s degree in computer science, mathematics, electrical engineering, or related fields;
• Fluent written and verbal communication skills in English;
• Ability to work both independently and collaboratively as part of a research team.

Desired background

We are looking for candidates with a strong background in one or more of the following areas:

• Stochastic processes, ideally including a background in ergodicity theory;
• Reinforcement learning, dynamic programming, and Markov decision processes;
• Programming skills (Python).

What we offer

• Fully funded doctoral researcher’s position at Aalto University, which is consistently ranked among the top universities in Europe;
• The position is fixed term and follows the school’s standard 2+2 model. It will be made initially for two years, with a six-month probationary period, and extended by two further years after a successful mid-term review, giving a total duration of four years;
• The position starts in January 2027 or as mutually agreed;
• The starting salary for a doctoral researcher is 3143 €/month;
• Opportunity to work on a high-impact research project;
• You will join a young and dynamic research group with ample opportunities for collaboration and exchange of ideas. The group is well connected internationally, with active research ties to institutions including RWTH Aachen University, KTH Stockholm, and the London Mathematical Laboratory, among others, and there is potential for research visits to these and other partners over the course of the PhD.